On Tue, 2015-11-24 at 20:12 +0000, Ian Campbell wrote: > The serial logs begin at the start of the flight, which might be long > before the job actually started. (confusing, I agree)
We could switch this to collect logs only from the start of the job, rather than the start of the flight, but this would still include a potentially long time in the hosts-allocate phase, where several other jobs could run. I investigated other options a while back: I found that sympathy has no way to ask it to rotate the logs (either globally or for a specific host, which would allow us to cause the relevant job's logs at least start on a file break boundary) nor any mechanism to inject a string into the log (to fingerprint the start of a phase to make searching easier etc). I also investigated making Osstest::Serial::* filter based on time stamps, but it was getting pretty fiddly. Perhaps not impossible to achieve, but still suffers from not being able to know when to cut off (maybe hosts- allocate could set a runvar?) Perhaps one simple aid would be to have osstest-confirm-booted log the flight + job along with an easily searchable pattern (like how I can usually search for *** in a build log to see what went wrong)? Ian. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
